Accelerated Deconvolution microscopy based on iterative coordinate descent algorithm using GPU

Abstract

In biological research, optical sectioning microscopy is widely used to observe the 3D structures of biological cells and tissues. However, the measured sectioning images are obscured by the out-of-focus information. To reduce such out-of-focus interference, it is necessary to implement a deconvolution method. Therefore, we employed iterative coordinate descent(ICD) algorithm which is based on updating of the pixel to optimize the statistical cost function iteratively. Even if the ICD method has rapidly convergent characteristic, the data complexity and the computational cost remain as limiting factors. To overcome this limitation, we apply a parallelized computing process occur in GPU hardware, known as general purpose computing on graphic processing units(GPGPU). In this thesis, we focus on implementation of the ICD algorithm based on the GPU architecture. From simulation data, we verify the performance by testing several realization of the proposed algorithm.